Closed
Description
What is the bug?
There is a small bug in the TextEmbeddingModel
class in relation with the new asymmetric model support. In the method addPrefixesToData
, that is only called when an asymmetric model is used, I forgot to add the existing result filter to the new instance of the TextDocsInputDataSet
that is created with the corresponding prefixes.
I noticed the issue while working on opensearch-project/neural-search#710.
How can one reproduce the bug?
Steps to reproduce the behavior:
- run inference on an asymmetric model using a result filter
- see that the result filter will not be applied
What is the expected behavior?
The result filter should be applied.
What is your host/environment?
- OS: all
Do you have any screenshots?
N.a.
Do you have any additional context?
I have a fix and will open a PR immediately.